Claremont CollegesScholarship @ Claremont

CMC Senior Theses CMC Student Scholarship

2018

Nonperforming Loans: Asset Pricing andDeterminants of ProfitabilityTyler Wheetley Amaya

This Open Access Senior Thesis is brought to you by Scholarship@Claremont. It has been accepted for inclusion in this collection by an authorizedadministrator. For more information, please contact scholarship@cuc.claremont.edu.

Recommended CitationWheetley Amaya, Tyler, "Nonperforming Loans: Asset Pricing and Determinants of Profitability" (2018). CMC Senior Theses. 1933.http://scholarship.claremont.edu/cmc_theses/1933

Claremont McKenna College

Non-Performing Loans: Asset Pricing and

Determinants of Profitability

Submitted to

Professor Batta

By

Tyler Wheetley Amaya

For

Senior Thesis

Spring 2018

March 28, 2018

1

I. Introduction

The decline in the quality of bank loan portfolios was a core problem

within the U.S. banking system during the recent financial crises (Kauko, 2011).

Mortgage delinquencies spiked, and the large amount of bad debt led to

stagnation in the economy (Sanders, 2008). Non-performing loans (NPLs) have

similarly been linked to other financial crises around the world, including the

crisis in the Eurozone (Maki, Tsagkanos, & Bellas, 2014) and the Asian financial

crisis (Yang, 2003). Researchers have investigated the negative correlation

between the number of non-performing loans (NPLs) in the economy and economic

conditions; finding that NPLs are generally more prevalent during times of

recession (Makri, Tsagkanos & Bellas, 2014). There are a variety of studies

focused on NPLs on a macroeconomic scale, as noted above. To the best of my

knowledge, however, there exists very little information on privately held NPLs

outside the banking industry.

The purpose of my paper is to add to the existing literature by examining

whether certain NPL loan characteristics (such as, the current interest rate,

delinquent taxes, current interest and principle payment, current taxes and

insurance payment, days in foreclosure, reason for default, occupancy status,

property condition, neighborhood condition, liquidation type, days held, UPB,

Acquisition BPO and state) lead to more profitable outcomes when being purchased

by non-banking firms. I am interested in assessing the accuracy of the current

asset pricing model for NPLs, specifically from the buy side. In particular, I

am testing whether certain NPL loan characteristics, some of which are not

directly controlled for in the current NPL pricing model, play a significant

role in predicting profitability. The loan characteristics used in the current

asset pricing model (such as, sponsor value, UPB, current interest rate and

occupancy status) are not expected to be significant in determining profit

2

outcomes as they have already been accounted for in determining the purchase

price of the asset (See Table 5 for current asset pricing model inputs).

Using private company data, I use a standard linear regression model to

examine the role of loan characteristics on NPL profitability. Intuitively, I

find that properties in good and fair condition positively affect NPL

profitability, relative to properties in poor or unknown condition.

Surprisingly, I find that interest rates are negatively correlated with NPL

profitability. This was unexpected because interest rates are a factor used to

determine the purchase price in the current asset pricing model. I believe this

outcome is due to interest rates being correlated with the probability of

reperformance. Due to the particular strategy of the firm used in this study,

NPLs which reperform are less profitable. I find a negative correlation with

NPL profitability and properties located in New York. This is likely because

New York is relatively more expensive than other states. I also find a very

slight correlation with Delinquent Taxes and profitability. This outcome can be

explained by the amount of property taxes being connected to property value,

where assets with a larger amount of delinquent taxes may be more likely to

have a higher property value.

The remainder of this paper is organized as follows. Section 2 presents

a background of the existing literature relating to NPLs on the macroeconomic

level. Section 3 presents the existing literature relating to housing prices

and variation of state foreclosure and liquidation laws. The data are described

in Section 4. Section 5 discusses the methodology and presents the results. The

final section concludes.

II. Background

The existing research on non-performing loans (NPLs) focuses mainly on

their relationship with banking practices and/or macroeconomic correlations.

3

Some commonly used macroeconomic indicators include, GDP and unemployment.

Curak, Pepur and Poposki (2013) look at determinants of NPLs, focusing on

Southeastern Europe and find that both macroeconomic conditions and banking

practices influence the number of non-performing loans in an economy. For NPLs,

they find a negative relationship with GDP and bank size. They also find a

positive relationship with inflation, real interest rates, and the solvency

ratio. Macroeconomically, this suggests that NPLs become less prevalent as GDP

increases and become more prevalent as inflation and interest rates increase.

This also suggests that an individual bank will hold fewer NPLs in their

portfolio as they grow larger and will hold a greater number of NPLs as their

solvency ratio increases. Klein (2013), with the IMF, finds similar results

when looking at Central, Eastern and Southeastern Europe (CESEE). He discovers

a feedback loop between NPLs and economic conditions, where the number of NPLs

in an economy will affect the state of the economy while economic conditions

will also affect the amount of NPLs. In general, research relating to

macroeconomic conditions finds that NPLs have a negative relationship with

preferable economic conditions, where the number of NPLs will increase during

times of recession.

Some banking practices have also been linked to the existence of NPLs.

Gosh (2015) evaluates banking practices and finds that bank profitability

reduces the number of NPLs a bank will hold. He also finds that liquidity risks

and poor credit quality increase NPL prevalence. This suggests that as banks

take on greater risks, such as those associated with credit and liquidity, they

will hold a larger number of NPLs in their portfolio. Looking at credit more

specifically, Ranjan and Chandra Dhal (2003) find that variables of credit

strongly affect bank-level NPAs (Non-performing assets) in India. Relating to

terms of credit variables, they find that expectations of increasing interest

rates will increase NPAs, while scope of credit maturity, healthier credit

4

culture, and better macroeconomic conditions reduce NPAs. This suggests that

credit related risks will increase as credit maturity is lengthened and as

credit culture diminishes. Intuitively, this also suggest that interest rates

and macroeconomic conditions affect not only the number of NPLs in an economy

but also the number of NPLs held by individual banks. In general, research

relating to banking practices mainly finds that NPLs have a positive

relationship with bank risk-taking behavior and a negative relationship with

bank size and profitability.

Only one paper I found looks at NPLs using non-bank data. Mahmood Rifat

(2016) focuses on the determinants of NPLs in the Non-Bank Financial

Institutions (NBFI) in Bangladesh. He finds that ROA differed significantly

between organizations. Similarly, the data I use also focuses on NPLs in the

private non-banking sector. However, I look more specifically at NPLs sold to

a private investment firm due to continued nonperformance. This firm targets

NPLs with a small probability of reperformance. There exits little to no

research in this area, especially given the specific strategy of this firm. I

intend to add to the literature by focusing on NPLs from the buy side and

looking at determinants of NPLs profitability compared to the current NPL model,

while focusing mainly on Midwestern and Southern U.S. states. This differs from

the current research because I look at privately held non-bank owned NPLs to

evaluate the financial processes involved from acquisition to liquidation. Based

on the structure and strategy of this firm I will not look at the average NPL

but instead those in significant distress and with little chance of re-

performance.

III. Literature Review

To my knowledge, there exists no literature on NPLs purchased by private

for-profit firms relating to NPL asset pricing or profit outcomes. Because

the literature in this area is scarce, I use exiting literature relating to

5

home valuation and state foreclosure, eviction and tax laws. Property

valuation and state law variation are two important factors in which I was

able to relate existing literature to NPL pricing and profitability.

As noted above, two significant NPL characteristics used in the current

NPL model of the firm are the sponsor value (similar to the acquisition BPO)

and the state in which the property is located. The sponsor value is an

estimate of the property value provided by the seller and the acquisition BPO

is an estimate of the property value provided by a broker around the time of

acquisition. The sponsor value is used as a preliminary value until the firm

can acquire its own value estimates from an outside appraiser. Property value

plays an inherently significant role in pricing an NPL and due to the unique

strategy of the firm whose data is used in my research, it is especially

significant. The firm targets NPLs with little change of reperformance which

typically results in foreclosure and REO sale. Thus, the underlying home

value is crucial in determining the purchase price. However, the purchase

price, or NPL Value, is capped by the unpaid principal balance (UPB) because

the UPB value is the maximum amount collectible in the case of reperformance.

The state in which the property is located also plays an especially important

role as it determines many associated timelines and costs affecting the NPL

purchase price value.

Many pieces come into play when valuing a residential property. Home

prices are affected by both internal and external characteristics. Obviously,

home values are affected by their inherent characteristics, such as property

size, the number of bedrooms and bathrooms, the year it was built and quality

of upkeep, and additional home amenities. Location it another a key

consideration in real estate valuation. Specifically, Bitter and Krause

(2017) discuss the impact of neighborhoods on home values finding that

certain neighborhood “packages” have a significant impact on home values.

6

There are a variety of neighborhood attributes which can contribute to home

value. Kane, Riegg and Staiger (2006) focus on school quality in

neighborhoods, finding that home prices vary significantly by school

assignment zone boundaries. Housing prices also vary significantly by state

and region. For example, Quigley (2005) found that housing costs in

California are relatively high compared to other states. Similarly, home

values are affected by geographical conditions associated with the locational

variation mentioned above. Benson (1998) focuses on the impact of ocean views

on home values. They find that the best ocean views increased home value by

nearly 60% while even low-quality ocean views could increase home value by

about 8%. Weather is another locational and geographical contributor to

property values. Harrison, Smersh & Schwartz (2001) find that homes located

within specific flood zones in Florida are on average worth less than

comparable homes outside the flood zones.

Other external conditions also affect home prices. Intuitively, the

state of the housing market will affect home prices (Quigley, 1999).

According to Davis and Nieuwerburgh (2014), housing also has a “lead-lag”

relationship with the business cycle. Thus, home prices will be lower

directly preceding and during times of recession. Borio and McGuire (2004)

note that traditional macroeconomic determinants of housing prices include

interest rates, output growth and unemployment. They also find that housing

prices often trend with equity prices and note that real home prices are

affected by inflation.

State laws are another important factor in determining the costs of

foreclosure, eviction and liquidation. These factors are included in the

current pricing model for NPLs (See table 6 for current NPL model state

assumptions). The variation in these laws is therefore also significant in

determining the value of an NPL. Terrence and Herzog (1990), note that these

7

costs can be categorized into three areas: transaction costs, property costs,

and opportunity costs. They define transaction costs as costs associated with

foreclose and liquidation, such as attorneys’ fees, brokers’ commissions and

title charges. Property taxes, insurance, utilities, repairs and property

maintenance are provided as examples of property costs. Foregone interest is

given as an example of an opportunity cost. Terrence and Herzog (1990) also

state three additional major differences in foreclosure laws between states:

1) Judicial vs nonjudicial procedure 2) the provision for a statutory right

of redemption and 3) deficiency judgements. They note that sates in which the

lender must use a judicial procedure to foreclose are costlier to firms

because the process tends to take longer. Conversely, states allowing

nonjudicial foreclosure, in which the lender can foreclose without going

through the court system, tend to be quicker, and thus less costly. Statutory

redemption relates to a borrower’s right to recover their ownership of a

property which has been foreclosed (Terrence and Herzog, 1990). States which

give borrowers this right can be costlier to the firm by delaying eviction

and extending asset holding times. Extending holding timelines increases

costs because the firm must pay certain property expenses periodically, such

as utilities and maintenance. A deficiency judgment is a ruling made by a

court in favor of the lender if a borrower’s mortgage foreclosure sale does

not produce sufficient funds to pay the loan in full (Terrence and Herzog,

1990). Because the firm caps their bid price by the UPB, deficiency

judgements are less important in this case. These state-specific foreclosure

laws are significant in calculating time and transaction related costs

associated with a given NPL.

IV. Data

The dataset used in the analysis is from a private asset management

firm, Neighborhood Stabilization Capital Management (NSCM). This dataset was

8

chosen because it provides loan level details for all assets purchased by the

firm (for example, loan ID, occupancy status, property condition, bankruptcy

status, liquidation amount, liquidation date, reason for default, current

interest rate, workout option, days in foreclosure, liquidation type,

property address, investor pool, entity, deal name, funding date, unpaid

principle balance (UPB), acquisition broker price opinion (BPO), purchase

price, and sum of total remittance). This data was retrieved from the firm’s

online database and remittance report. The workout option data has been

excluded because it is collinear with liquidation type. The sum of total

remittance was used as a reference but was excluded from the model because

remittance data was captured through the IRR calculation. Assets currently

under management have been excluded due to the limited information available

for non-liquidated assets and to maintain a complete dataset. Besides the

remittance data used in the IRR calculations, any data obtained after the

time of acquisition was excluded for the purpose of evaluating the current

NPL pricing model.

The final sample includes 179 assets (that is, mortgage-backed notes

and REO properties) both purchased and liquidated between 2014 and 2018. The

majority of the sample includes properties from Midwestern, Southeastern and

Northeastern states. The sample is comprised of 48 properties located in

Illinois, 36 properties in Georgia, 26 properties in New Jersey and 16

properties in Florida, 9 properties in Maryland, 7 properties in

Pennsylvania, 6 properties in Ohio, 6 properties in New York, 6 properties in

New England states, and 19 properties in other states. All states included in

the other category contain fewer than 5 assets. As such, these remaining

states are aggregated into a single category (See Table 4, Panel A for asset

breakdown by state).

9

The internal rate of return (IRR) of each loan is used as a measure of

profitability and is calculated using the total loan remittance over time.

Days held is calculated as the difference between the Funding Date and the

Liquidation Date. Sum of total remittance is calculated as the net remittance

amount, excluding time. The liquidation amount provided is net of expenses

and is used in the IRR calculation through the remittance report. Property

condition and Neighborhood condition data was acquired from the servicers at

the time of acquisition. Explanations for individual dummy variables are

provided below. All other variables were exported directly from the database

and are unadjusted (See Table 1 for variable definitions).

Each asset in the sample is categorized as belonging to one of two

mutually exclusive default reasons: Default reason and other. Default reason

is used when no reason is give and is aggregated with assets in which the

reasons given include: Other/No Applicable Code and Unable to Contact

Borrower. Other default reasons include, temporary Loss or Reduction of

income, unemployment, death or illness of Primary Borrower, excessive

obligations, servicing problems, and marital difficulties. I define these

variables using information provided by the servicers at the time of

acquisition. For each category, I assign a value of one if the asset falls

within the category, and zero otherwise. Majority of borrowers did not give a

reason for default. 152 of the 179 assets fall within the Default/Other

category meaning this category is limited in the information is can provide.

Only 27 of the 179 borrowers provided a reason for default. (See Table 4,

Panel B for asset breakdown by default reason).

Assets are categorized as belonging to one of three mutually exclusive

Occupancy Statuses: Vacant, Owner Occupied, and Other/Unknown. These

variables are defined by who was living in the home at the time of

acquisition. For each category, I assign a value of one if the asset falls

10

within the category, and zero otherwise. Non-owner-occupied assets are

aggregated with other/unknown. 89 of the 179 assets fall under vacant, 74 of

the assets fall under owner occupied and 16 of the assets fall under

Other/Unknown. (See Table 4, Panel C for asset breakdown by occupancy status)

Assets are categorized as belonging to one of two mutually exclusive

bankruptcy statuses: BK and None. I define these variables by if the borrower

declared bankruptcy while the firm held the loan. For each category, I assign

a value of one if the asset falls within the category, and zero otherwise.

Majority of the assets fall within the None category in which case the

borrower did not declare bankruptcy. 43 of the 179 assets went into

bankruptcy. The BK status includes chapters 11, 7 and 13. (See Table 4, Panel

D for asset breakdown by bankruptcy status)

Assets are categorized as belonging to one of three mutually exclusive

Property Conditions: Good, Fair, and Poor/Unknown. I define these variables

using information provided by the servicers at the time of acquisition. For

each category, I assign a value of one if the asset falls within the

category, and zero otherwise. 93 of the 179 assets were in good condition at

the time of acquisition, 51 were in fair condition and 16 were either in poor

condition or the condition was unknown (see Table 4, Panel E for asset

breakdown by property condition). Poor and unknown were aggregated because

they are the smallest categories, however, this category may provide limited

information due to assets with unknown property conditions diluting the

results of the assets in poor condition. Assets are similarly categorized as

belonging to one of two mutually exclusive Neighborhood Conditions: Stable

and Declining/Unknown. For each category, I assign a value of one if the

asset falls within the category, and zero otherwise. These variables are also

defined using information provided by the servicers at the time of

acquisition. 139 of the 179 properties were located in stable neighborhoods

11

at the time of acquisition and 40 of the 179 were in either declining

neighborhoods or the neighborhood condition was unknown (See Table 4, Panel F

for asset breakdown by neighborhood condition). Declining and unknown were

also aggregated because they are the smallest categories. Similarly, this

category may also provide limited information due to assets with unknown

neighborhood conditions being included with the assets in poor condition.

Each asset in the sample is categorized as belonging to one of two

mutually exclusive liquidation types: REO sale and other liquidation type. I

define these variables by how the asset was sold. For REO, I assign a value

of one if the asset was sold as an REO, and zero otherwise. Similarly, for

other liquidation type, I assign a value of one if the asset was not sold as

an REO, and zero otherwise. Other liquidation types include, note sales,

third party sales, short sales, foreclosure sales, not listed, and one which

was put back to the seller. 81 of the 179 assets were liquidated through REO

sales and 71 of the 179 assets were liquidated in another form (See Table 4,

Panel G for asset breakdown by state).

Assets are categorized as belonging to one of three mutually exclusive

entities: Southside NSP Trust 2016-1, Southside NSP Trust 2015-1, and Private

Trust 2014-1. I define these variables by how and when the assets were

funded. For each category, I assign a value of one if the asset falls within

the category, and zero otherwise. 27 of the 179 assets fall under Southside

NS Trust 2016-1, 79 fall under Southside NSP Trust 2015-1 and 73 fall under

Private Trust 2014-1. (See Table 4, Panel H for asset breakdown by entity)

Assets were categorized as belonging to one of five mutually exclusive

Workout Options: Short Sale, Repayment Plan, Short Repay Plan, Modification,

None/Other. I had defined these variables by weather and how a borrower

became current on their mortgage. For each category, I assigned a value of

one if the asset fell within the category, and zero otherwise. Majority of

12

the assets fell within the none/Other category in which case an alternative

work out option was used or, pertaining to majority of the assets within this

category, there was no workout option and the home was foreclosed. This

category was omitted because it is perfectly collinear with the liquidations

types category. This is intuitive because certain workout options correlate

with specific liquidation types. For example, if there was a loan

modification, then the asset will be liquidated as a note sale. Similarly, if

the asset had no workout option then it would be liquidated though an REO

sale. Thus, this information is captured by the Liquidation Types category.

V. Determinants of NPL Profitability and Purchase Price

To formally identity the underlying causes of differences in NPL

profitability, I perform a standard linear regression of the determinants of

NPL pricing characteristics. Specifically, I estimate a model of NPL

profitability of the following form:

(1) Yi = Xi1β1 +εi

where Yi is the internal rate of return (IRR) which captures NPL

profitability and i represents individual NPLs. X is a vector of observable

characteristics, and 𝜀 is an error term with the usual properties. To provide

a base comparison of what has already been captured in the asset pricing

model, I re-estimate equation (1) by replacing the NPL measure of

profitability with the NPL purchase price. This provides an understanding of

the extent that each loan characteristic contributed to the initial price,

under the current asset pricing model.

Unsurprisingly, the majority of the loan characteristics are statistically

insignificant. This suggests that either the variation in those loan

characteristics have been adequately accounted for in the current asset

13

pricing model or the characteristics do not consistently contribute to

specific profit outcomes. This is particularly unsurprising for the state,

occupancy status, and Acquisition BPO variables because those are the primary

characteristics used in the asset pricing model to determine the purchase

price of the assets (See tables 2 and 3 for summary statistics and results).

Interestingly, the current interest rate is statistically significant and

negatively correlated with IRR, with a coefficient of -45.48. Interest rates

are considered in the current asset pricing model and they also have a

negative correlation with the purchase price. However, this implies that an

increase in the interest rate will reduce the firms calculated purchase price

but will still result in a lower profit outcome. Thus, the variation in the

interest rate may not be adequately considered in the current NPL pricing

model. This suggests that interest rates should be given greater

consideration in the determination of the purchase price. According to the

results, the firm should reduce its purchase price by a greater amount than

it currently does with an increase in the interest rate to counterbalance the

lower profit outcome.

Delinquent taxes is statistically significant in contributing to

profitability but has only a slightly positive correlation with IRR, with a

coefficient of .00018. It is not statistically relevant to the Purchase Price

as they are deducted directly from the purchase price. This suggests that

delinquent taxes signify something slightly greater than their numerical

amount. It is possible that this is a result of higher valued properties

having a larger amount of property taxes. The acquisition BPO should capture

the variation in property value, however, BPOs can vary significantly and it

difficult to pinpoint the exact value of a given property.

New York is statistically significant and negatively correlated with IRR

with a coefficient of -11.546. This is likely because New York is relatively

14

more expensive than other states. This suggests that the firms model does not

accurately account for the additional costs associated with properties in New

York. According to the results, the firm should assign a lower purchase price

than it currently does to properties in New York. However, only 6 of the 179

assets are located in New York, making this a very small sample size.

Additional data on properties in New York should be included to better asses

this variable.

Intuitively, properties in good and fair condition are both positively

correlated with IRR, relative to properties in poor or unknown condition.

They have coefficients of 3.13 and 2.53, respectively. Although intuitive,

these results are still somewhat surprising as the variation in property

condition should be captured though the acquisition BPO. However, BPO

accuracy is limited by the significant room for error in property valuation.

Unsurprisingly, the R2 is significantly smaller for IRR than for

Purchase Price; 0.372 and 0.965, respectively. This was expected because the

asset characteristics used in the model directly reflect the characteristics

used in the asset pricing model to determine the purchase price. Intuitively,

the loan characteristics considered in the model more adequately reflect the

current pricing model than the true profit outcomes. Thus, additional

variables would need to be included to more accurately predict profit levels.

However, limiting the model to information available at the time of

acquisition was expected to reduce the R2.

Including additional NPL characteristics could create a more

comprehensive model which would better predict profitability. For example,

the variation in the BPOs of a given asset over time would likely be an

important predictor of profit outcomes because the variation reflects the

accuracy of those value estimates. If a property has BPOs that vary

significantly, it is more likely that the value attributed to that asset

15

could be inaccurate. A property with a small variation in BPO values is more

likely to have an accurate value attributed to it. Although significant, this

information is unimportant in relation to the asset pricing model because an

asset must be priced at the time it is acquired. BPOs are ordered

periodically during the time an asset is held, making this information

inaccessible at the time of acquisition. Thus, for the purpose of evaluating

the current pricing model, information acquired after the time of acquisition

was excluded from my model.

VI. Conclusion

I formally analyze the role of NPL loan characteristics in explaining NPL

profit outcomes compared to the current pricing model for NPLs. To the best

of my knowledge, this topic has not been examined in any previous studies.

Existing literature looks largely at determinants of NPL prevalence in the

economy and their relationship with banking practices. I am contributing the

literature by evaluating determinants of NPL profitability within a private

asset management firm. I use the internal rate of return (IRR) of each loan

as a measure of profitability. I expected that factors included in the

current NPL model would not be statistically significant in determining

profit outcomes as those factors were considered in determining the purchase

price of the asset.

Surprisingly, I find that interest rates are statistically significant and

negatively correlated with IRR. This is surprising because interest rates are

considered in the current asset pricing model. The results suggest that

greater weight should be given to interest rates in determining the purchase

price of an NPL. Also surprising, I find that properties located in New York

are negatively correlated with profitability. State assumption relating to

costs and holding timelines are included in the current NPL model. However,

the results suggest that additional consideration should be given to

16

properties in New York when determining a purchase price. I also find that

delinquent taxes have a slightly positive correlation with NPL Profitability.

This was unexpected as delinquent taxes are deducted directly from the

purchase price, suggesting a slightly greater significant than their dollar

amount.

The NPL loan characteristics used in the model explain less than half the

variation in IRR, suggesting that additional factors should be considered to

better represent the total variation in profit outcomes. However, for the

purpose of evaluating the current pricing model, I excluded any data acquired

after the time of acquisition. Including additional NPL information attained

after acquisition may be useful in creating a more comprehensive model of the

determinants of NPL profitability, however, this would be unhelpful in

contributing to predictors of the asset pricing model. When using only

information available at the time of acquisition, that information is

limited. Even if it were possible for the firm to retrieve additional

information on each asset prior to acquisition, retrieving additional

information may be costly. Asset information may have marginal benefits and

the costs of retrieving additional data may outweigh those benefits.

The results are also limited by the relatively small sample size of the

data and the significant variation in IRR. Additional research should be

performed with a larger sample size to more accurately analyze the impact of

loan attributes on profitability, particularly on properties in New York. The

variation in outcomes may be a result of both the size and youth of the firm.

In particular, two pools of assets retrieved inaccurate BPOs resulting in

very negative profit outcomes for those assets. This occurred during the

early years of the firm’s life and operations have since been improved to

prevent future BPO inaccuracy. It is possible that assets with bad BPOs could

have skewed some results. Using data from a larger more mature company may

17

provide more precision in the results. With the small sample size of the data

I was not able to exclude the assets with bad BPOs or some other potential

outliers. Obtaining a larger sample size would likely be the best possible

improvement to this study.

18

Table 1: Variable Definitions

Variable Definition

IRR The rate of return that sets the NPV of all cash flows

(both positive and negative) from the investment to ZERO.

Individual asset IRRs are calculated from their total

remittance.

Current

Interest Rate

The proportion of the loan charged as interest to the

borrow, expressed as an annual percentage of the loan

outstanding.

Delinquent

Taxes

The amount of property taxes unpaid after the payment due

date.

Current

Principle and

Interest

Payment

The current payment amount towards the principle owed and

the amount determined by the interest rate.

Current Taxes

and Insurance

Payment

The payment toward the amount of principle owed and the

amount determined by the interest rate

Days in FC The number of days the asset was in foreclosure.

Default Reason A dummy variable equal to one if no default reason was

given by the borrower or the reason is otherwise unknown.

Other (omitted) A dummy variable equal to one if a default reason was

provided by the borrower.

Owner Occupied A dummy variable equal to one if the property owner lived

in the home at the time of acquisition.

Other/Unknown A dummy variable equal to one if someone other than the

property owner lived in the home at the time of acquisition

or if the occupancy status was unknown.

Vacant

(omitted)

A dummy variable equal to one if the property was vacant at

the time of acquisition.

Good A dummy variable equal to one if the property was in good

condition at the time of acquisition.

Fair A dummy variable equal to one if the property was in fair

condition at the time of acquisition.

Poor/Unknown

(omitted)

A dummy variable equal to one if the property was in good

condition at the time of acquisition.

Bankruptcy (BK) A dummy variable equal to one if the borrower declared

bankruptcy.

None (omitted) A dummy variable equal to one if the borrower did not

declare bankruptcy.

Stable A dummy variable equal to one if the neighborhood was

stable at the time of acquisition.

Declining/

Unknown

(omitted)

A dummy variable equal to one if the neighborhood was

declining at the time of acquisition or if the neighborhood

condition was unknown.

REO Sale A dummy variable equal to one if the asset was liquidated

through and REO sale.

Other

Liquidation

Type (omitted)

A dummy variable equal to one if the asset was not

liquidated through and REO sale.

SouthsideNSP

Trust 2016

A dummy variable equal to one if the asset were funded by

the Southside NSP Trust 2016.

SouthsideNSP

Trust 2015

A dummy variable equal to one if the asset were funded by

the Southside NSP Trust 2015.

Private Trust

2014 (omitted)

A dummy variable equal to one if the asset were funded by

the Private Trust 2014.

19

Investor Pool Indicator of the investor pool in which the asset falls

under.

Deal Name Indicator of the deal name in which the asset falls under.

Days Held The number of days the asset was held, calculated as the

difference between the funding date and liquidation date.

UPB The principle amount still owed on the loan.

AcqBPO The broker price opinion obtained at the time of

acquisition.

Zip Address zip code.

GA (omitted) A dummy variable equal to one if the property is located in

Georgia.

NJ A dummy variable equal to one if the property is located in

New Jersey.

FL A dummy variable equal to one if the property is located in

Florida.

IL A dummy variable equal to one if the property is located in

Illinois.

NY A dummy variable equal to one if the property is located in

New York.

MD A dummy variable equal to one if the property is located in

Maryland.

OH A dummy variable equal to one if the property is located in

Ohio.

PA A dummy variable equal to one if the property is located in

Pennsylvania.

New England A dummy variable equal to one if the property is located in

a New England state.

Other States A dummy variable equal to one if the property was not

located in Georgia, New Jersey, Florida, Illinois, New

York, Maryland, Ohio, Pennsylvania, or any New England

States.

20

Table 2: Summary Statistics This table presents the summary statistics for the variables used

in our study. The sample period is from 2015 to 2018. The definitions of the variables are provided in

Table 1.

mean sd p25 p50 p75

IRR

0.672 4.450 -0.015 0.014 0.039

Purchase Price 67658.490 57129.970 31436.220 57271.990 80644.560

Current Interest Rate 0.058 0.020 0.050 0.058 0.065

Delinquent Taxes 1387.342 6455.685 0.000 0.000 1046.000

Current Principle and

Interest Payment 3556.867 36975.460 465.560 679.250 932.920

Current Taxes and Insurance

Payment 291.016 245.785 120.590 249.920 440.240

Days in Foreclosure (FC) 278.760 442.935 0.000 43.000 441.000

Default Reason 1.611 10.754 1.000 1.000 1.000

Other (omitted)

Owner Occupied 0.822 5.507 0.000 0.000 1.000

Other/Unknown 0.178 1.220 0.000 0.000 0.000

Vacant (omitted)

Good 1.033 6.911 0.000 1.000 1.000

Fair 0.567 3.807 0.000 0.000 1.000

Poor/Unknown (omitted)

Bankruptcy (BK) 0.478 3.216 0.000 0.000 0.000

None (omitted)

Stable 1.544 10.311 1.000 1.000 1.000

Declining/Unknown (omitted)

REO Sale 1.200 8.020 0.000 1.000 1.000

Other Liquidation Type

(omitted)

SouthsideNSP Trust 2016 0.300 2.033 0.000 0.000 0.000

Southside NSP Trust 2015 0.878 5.876 0.000 0.000 1.000

Private Trust 2014 (omitted)

Investor Pool 7.615 4.364 3.000 8.000 10.000

Deal Name 15.061 20.672 3.000 8.000 19.000

Days Held 431.374 237.703 233.000 357.000 583.000

Unpaid Principle Balance

(UPB) 128795.400 109628.600 59859.470 105775.800 153863.600

Acquisition BPO (AcqBPO) 117589.900 91123.110 63500.000 97500.000 142000.000

GA (omitted)

NJ 0.289 1.959 0.000 0.000 0.000

FL 0.178 1.220 0.000 0.000 0.000

IL 0.533 3.585 0.000 0.000 1.000

NY 0.067 0.480 0.000 0.000 0.000

21

MD 0.100 0.702 0.000 0.000 0.000

OH 0.067 0.480 0.000 0.000 0.000

PA 0.078 0.554 0.000 0.000 0.000

New England 0.067 0.480 0.000 0.000 0.000

Other States 0.211 1.442 0.000 0.000 0.000

22

Table 3: NPL Loan Characteristics: Linear Regression Estimations

This table presents the results of regressions of NPL Loan

characteristics on IRR and Purchase Price. All variables are

defined in Table 1. t-statistics are in parentheses. Significant at

the 10%, 5%, and 1% levels is indicated by *,**,***, respectively.

IRR Purchase Price

CurrentInterestRate -45.47803** -161316.6***

(-2.39) (-2.71)

DelinquentTaxes 0.0001859 -0.091278***

(3.11) (-0.49)

PrincipleandInterestP -5.71e-06*** -0.0756715

(-0.62) (-2.63)

TaxesandInsurancePaym 0.0000852 -4.500637

(0.05) (-0.90)

DaysinFC -0.0002725 -4.114386*

(-0.37) (-1.80)

DefaultReason -1.198282 -2103.248

(-1.39) (0.78)

OwnerOccupied -0.6255507 -1310.846

(-0.71) (-0.48)

OtherUnkown -0.3628081 1380.225

(-0.28) (0.33)

Good 3.132552*** -30.01377

(2.69) (-0.01)

Fair 2.526089** 1249.455

(2.13) (0.34)

BK 0.7198147 -2622.612

(0.95) (-1.10)

Stable -1.833429 -3252.126

(-1.43) (-0.81)

REOSale -1.395161* -332.3978

(-1.84) (-0.14)

SouthsideNSPTrust20151 2.726646* -411.9387

(1.83) (-0.09)

SOUTHSIDENSPTRUST20161 0.3819479 -2802.783

(0.13) (-0.32)

InvestorPool -0.4316824 963.7667

(-1.41) (1.01)

DealName 0.076585 -237.4168

(1.06) (-1.05)

DaysHeld -0.0029275* 1.329684

(-1.78) (0.26)

UPB 1.09e-06 -0.0099328

23

(0.21) (-0.61)

AcqBPO 4.36e-06 .6352212***

(0.70) (32.51)

NJ 0.4943491 -4470.303

(0.38) (-1.11)

FL 0.4313081 3661.242

(0.36) (0.97)

IL -0.9088358 -1373.376

(-0.94) (-0.45)

NY -11.49263*** 2513.056

(-5.36) (0.37)

MD -2.134068 -226.9682

(-1.34) (-0.05)

OH 0.3481596 399.0185

(0.19) (0.07)

PA 2.097751 -6104.588

(1.28) (-1.19)

NewEngland -0.0756457 -7063.842

(-0.04) (-1.05)

OtherStates 1.978419 2456.018

(1.44) (0.57)

Constant 4.629093 8754.932

(1.76) (1.06)

Obs. 179 179

R2 0.4147 0.9652

Adj. R2 0.3008 0.9584

24

Table 4: Asset Breakdowns

These tables present the breakdown of each dummy variable by possible

aggregated subcategories and number of assets.

Table 4, Panel A:

Asset breakdown by

State

1.FL 16

2.GA 36

3.IL 48

4.MD 9

5.NJ 26

6.NY 6

7.OH 6

8.PA 7

9.New England: 6

CT (3)

MA (2)

VT (1)

10.Other States: 19

CO (2)

CT (3)

IA (2)

IN (2)

MN (2)

NC (4)

TX (5)

Total 179

0 20 40 60

Other States

New England

PA

OH

NY

NJ

MD

IL

GA

FL

Assets by State

Number of Assets

25

Table 4, Panel B: Asset breakdown by

Default Reason

1.Default Reason 149

Default Reason (140)

Unable to Contact Borrower (7)

Other/NO Applicable Codes (2)

2.Other 30

Temporary Loss/Reduction of

Income

(7)

Death/Illness of Primary

Borrower

(4)

Excessive Obligations (11)

Unemployment (3)

Servicing Problems (2)

Payment Dispute/Due Date (2)

Marital Difficulties (1)

Total (179)

Table 4, Panel C: Asset Breakdown

by Occupancy Status

1.Vacant 89

2.Owner Occupied 74

3.Other/Unknown 16

Non-Owner Occupied (7)

Other/Unknown (9)

Total 179

Table 4, Panel D: Asset Breakdown

by Bankruptcy Status

1.None 136

2.BK Chapter 43

BK 7 (24)

BK 11 (1)

BK 13 (18)

Total 179

0 50 100 150 200

Other

Default Reason

Assets by Default Reason

Number of Assets

0 20 40 60 80 100

Other/Unknown

Owner Occupied

Vacant

Assets by Occupancy Status

Number of Assets

0 20 40 60 80 100

Other/Unknown

Owner Occupied

Vacant

Assets by Occupancy Status

Number of Assets

26

Table 4, Panel E: Asset Breakdown by

Property Condition

1.Good 93

2.Fair 51

3.Poor/Unknown 35

Poor (7)

Unknown (28)

Total 179

Table 4, Panel F: Asset Breakdown by

Neighborhood Condition

1.Stable 139

2.Declining/Unknown 40

Declining (7)

Unknown (33)

Total 179

Table 4, Panel G: Asset Breakdown by

Liquidation Type

1.REO Sale 108

2.Other Liquidation Type 71

Note Sale (27)

Third Party Sale (20)

Short Sale/Short Payoff (3)

Foreclosure Sale (1)

Not Listed (19)

Put Back to Seller (1)

Total 179

0 20 40 60 80 100

Poor/Unknown

Fair

Good

Assets by Property Condition

Number of Assets

0 50 100 150

Declining/Unknown

Stable

Assets by Neighborhood Condition

Number of Assets

0 50 100 150

Other Liquidation

Type

REO Sale

Assets by Bankruptcy Status

Number of Assets

27

Table 4, Panel H: Asset

Breakdown by Entity

SOUTHSIDE NSP TRUST 2016-1 27

Southside NSP Trust 2015-1 79

Private Trust 2014-1 73

Total 179

0 20 40 60 80 100

SOUTHSIDE NSP TRUST

2016-1

Southside NSP Trust

2015-1

SouthsideE NSP Trust

2016-1

Assets by Entity

Number of Assets

28

Table 5: Asset Pricing Model Inputs

This table presents the factors used in the current NPL asset pricing model

for NS Capital. These loan characteristics are used to determine the

purchase price.

Loan

Characteristic

Definition

Current Status Status Options:

30-59 Days (delinquent)

60- 89 Days (delinquent)

90 + Days (delinquent)

UPB Unpaid Principle balance

Sponsor Value Estimated property value provided by seller (used as

preliminary estimate prior to acquisition BPO)

Next Payment

Due Date

Date of next borrower payment

Interest Rate Interest rate on mortgage

Occupancy

Status

Status Options:

Owner Occupied

Vacant

Other/Unknown

29

Table 6: Current NPL Asset Pricing Model State Assumptions

These tables present the state assumptions for all states included in the

dataset. These assumptions are incorporated into the asset pricing model to

determine the purchase price.

Table 6, Panel A: Eviction Assumptions

State Evic

Timeline

Evict

Costs

EvictConf-

Redemp

Evict

Most

Frequent

EvictNotes

IA 45 850 None

IL 219.5 1200 Confirmation 30-45

days

Also, a 30 day right

to possession once the

sale confirms before

you can start eviction

IN 40 850 - Up to 1-year

redemption if

mortgagee pursues a

deficiency judgment or

lender forecloses with

redemption

LA 120 750 - None

MA 75 750 - None

MD 195 1200 Ratification 45-60

days

Cannot begin marketing

until judge has signed

order of ratification

MN 57 950 Redemption 6

months.

If

vacant,

can

shorten

1 year if over 10

acres OR if debt is

less than 2/3 original

mortgage

NC 35 750 Confirmation 10 days Additional 10 day

upset bid period

NJ 140 1200 Redemption 10 days None

NY 172.5 1200 - None

OH 103.5 1000 Confirmation 45-60

days

None

PA 110 1000 None. Note: cannot

begin eviction until

deed has been recorded

TX 55 850 - None

VT 40 850 Redemption 6 months Can be shortened if

there is no equity in

the property

30

Table 6, Panel B: Foreclosure Assumptions

State Fcl

Judicial

Fcl Non-

Judicial

Fcl

Comment

Fcl

Process

Period

FCL

Timeline

Fcl

Redemption

FCL

Costs

CO • • Judicial

rarely

145 191 800

CT • Judicial

only

62 280 1500

FL • Judicial

only

135 390 750

GA • • Judicial

rarely

37 180 2750

IA • • Trustee

Sale

Voluntary

160 205 20 600

IL • Judicial

only

300 472 90 5250

IN • Judicial

only

261 296 1000

LA • Judicial

only

180 540 1150

MA • Judicial

only

75 338 1300

MD • Judicial

only

46 215 5250

MN • • Non-

Judicial

mostly

100 310 180 2000

NC • • Non-

Judicial

mostly

110 150 1500

NJ • Judicial

only

270 455 10 5250

NY • Judicial

only

445 1050 5250

OH • Judicial

only

217 203 2800

PA • Judicial

only

270 630 4250

TX • • Non-

Judicial

mostly

27 70 850

VT • Judicial

only

95 390 240 1700

31

Table 6, Panel C: REO, Tax and HPI Assumptions

REO

Tax HPI

State REO

Months

REO Days REO State

Speed

Tax Rate HPI - 2015 HPI - 2016

CO 6 90 2 1.08% 360.5 392.8

CT 8 120 3 1.72% 100.0 102.5

FL 5 120 4 1.20% 100.0 102.5

GA 6 181.5 3 1.52% 100.0 102.5

IA 8 90 3 2.15% 252.7 258.6

IL 6 181.5 4 2.50% 100.0 102.5

IN 6 90 2 2.12% 244.8 251.1

LA 6 120 3 1.02% 100.0 102.5

MA 8 90 3 1.07% 621.7 652.4

MD 6 181.5 4 3.00% 100.0 102.5

MN 6 120 2 1.27% 309.2 325.0

NC 8 120 3 1.10% 302.5 311.8

NJ 6 181.5 2 3.76% 100.0 102.5

NY 6 181.5 4 3.76% 100.0 102.5

OH 8 242 3 1.81% 100.0 102.5

PA 6 181.5 3 1.70% 100.0 102.5

TX 6 120 1 2.57% 234.8 253.0

VT 9 90 4 2.06% 436.2 439.0

32

Works Cited

Marijana Curak, Sandra Pepur and Klime Poposki (2013). Determinants of non-

performing loans – evidence from Southeastern European banking systems.

Banks and Bank Systems, 8(1).

Nir Klein (2013). Non-performing loans in CESEE: determinants and impact on

macroeconomic performance. IMF Working Papers.

Amit Gosh (2015). Banking-industry specific and regional economic

determinants of non-performing loans: evidence from US states. Journal

of Financial Stability. 20(1), 93-104.

Apel Mahmood Rifat (2016). An analytical study of determinants of non-

performing loans: evidence from non-bank financial institutions of

Bangladesh. JBT, 11(1-2).

Rolan Beck, Petr Jakubik and Anamaria Piloiu (2013). Non-performing loans:

what matters in addition to the economic cycle? Reserve Bank of India

Occasional Papers.

Karlo Kauko (2011). External deficits and non-Performing loans in the recent

financial crisis. Economic Letters, 115(2), 196-199.

Anthony Sanders (2008). The subprime crisis and its role in the financial

crisis. Journal of Housing Economics. 17(4), 254-261.

Makri, V., Tsagkanos, A., & Bellas, A. (2014). Determinants of non-performing

loans: The case of eurozone. Panoeconomicus, 61(2), 193-206.

33

Yang, L. (2003). The Asian financial crisis and non-performing loans:

evidence from commercial banks in Taiwan. International Journal of

Management, 20(1), 69-74.

http://ccl.idm.oclc.org/login?url=https://search.proquest.com/docview/2

33231338?accountid=10141

Christopher Bitter, Andy Krause, (2017). The influence of urban design

packages on home values. International Journal of Housing Markets and

Analysis, 10(2), 184-203. https://doi.org/10.1108/IJHMA-08-2015-0049

Thomas J. Kane & Stephanie K. Riegg & Douglas O. Staiger, (2006). School

quality, neighborhoods, and housing prices. American Law and Economics

Review, Oxford University Press, 8(2) 183-212.

Benson, E.D., Hansen, J.L., Schwartz, A.L. et al. The Journal of Real

Estate Finance and Economics (1998) 16: 55.

https://doi.org/10.1023/A:1007785315925

John M. Quigley (1999). Real estate prices and economic cycles. International

Real Estate Review. 2(1), 1-20.

David Harrison, Greg T. Smersh, and Arthur Schwartz (2001). Environmental

determinants of housing prices: the impact of flood zone status.

Journal of Real Estate Research: 2001, 21(1-2), 3-20.

Borio, Claudio E. V. and McGuire, Patrick, (2004). Twin peaks in equity and

housing prices? BIS Quarterly Review, March 2004.

https://ssrn.com/abstract=1491070

Terrence M. Clauretie and Thomas Herzog (1990). The effect of state

foreclosure laws on loan losses: evidence from the mortgage insurance

industry. Journal of Money, Credit and Banking 22(2) 221-233.

34

Quigley, John, M., and Steven Raphael, (2005). Regulation and the high cost

of housing in California. American Economic Review, 95(2), 323-328.